I am fairly new to RNAseq analysis and am wondering how to do analysis with a 3 factor treatment in DESeq2. I have plants infested with insect over days. My design is as follows: 2 genotype plants x 2 treatments (infested and control) x 3 days (plants sampled over 3 days).
I ran DESeq2 with the following design: design = ~ genotype + treatment + day
I now want to look at within days differences between genotype and treatments and would like to also do across days.
Could someone point me towards an example where people have looked across factors and within factors when conducting downstream contrasts?
How many replicates do you have per combination of genotype x treatment x day?
There's not enough information here for how to set up an analysis. Within day, you want to compare across genotype and across treatment? Do you believe that the genotypes have the same treatment effect? Do you think that the treatment effect is the same across day? When you want to compare across days, how so? Do you want to compare within groups of genotype x treatment across day? There are actually many ways to set up a design with three factors, and it's not trivial to interpret the results.
There are 3 replicates per combination, with a total of 36 experimental units . I expect genotypes to vary within days, one is resistant and the other is susceptible to insect treatment. I want to compare within days across genotype and treatment. But I also would like to look at differences among days within genotype and treatment. From running a pca I see there are differences among days. I think biologically it will make more sense to look within day at both the treatment and genotype effect.
This is a sufficiently complex experimental design that I think you should find a local statistical collaborator to help with interpretation of results. It sounds like you have reason to want to model interactions between all three factors, so you would use a design ~genotype * treatment * day. One way to justify this design is that, the treatment effect is genotype specific, and you expect that this difference itself could change across day. It's simple enough to specify the design but then you will need help with building contrasts to express the different comparisons you want to make. Note that the coefficients produced by DESeq2 with this design are the same ones you would encounter with any normal linear model, so you could approach anyone with a background in linear modeling to help you interpret the results. You can build various contrasts using numerous approaches outlined in the ?results function, under the contrast argument.
Hi Michael, I have been looking for an answer to a similar experiment as described above. You have been very helpful in answering lots of questions. I am stuck in a similar position. I have read the manuals and I was able to design models and analyse results based on a two factorial level. But having troubles generating results based on multi-factorial (and multi-level) experiments. My experiment is also the same as above and looking for answers to the same questions. It would be great if you could please suggest a similar answer as you have explained everything step wise in a different post for two-factor model design - DESeq2 Model Design . I'm sure lots of us will benefit from this - thanks
Hi, Did you get what you were looking for? It would be great if you could please share how you achieved your results. I'm exactly analysing a similar experiment and was looking for help in the forums. Specially with designing model and listing out contrasts for comparisons (treatment effect, genotype effect and treatment vs gentotype effect). thanks
There are 3 replicates per combination, with a total of 36 experimental units . I expect genotypes to vary within days, one is resistant and the other is susceptible to insect treatment. I want to compare within days across genotype and treatment. But I also would like to look at differences among days within genotype and treatment. From running a pca I see there are differences among days. I think biologically it will make more sense to look within day at both the treatment and genotype effect.
This is a sufficiently complex experimental design that I think you should find a local statistical collaborator to help with interpretation of results. It sounds like you have reason to want to model interactions between all three factors, so you would use a design
~genotype * treatment * day. One way to justify this design is that, the treatment effect is genotype specific, and you expect that this difference itself could change across day. It's simple enough to specify the design but then you will need help with building contrasts to express the different comparisons you want to make. Note that the coefficients produced by DESeq2 with this design are the same ones you would encounter with any normal linear model, so you could approach anyone with a background in linear modeling to help you interpret the results. You can build various contrasts using numerous approaches outlined in the ?results function, under the contrast argument.Hi Michael, I have been looking for an answer to a similar experiment as described above. You have been very helpful in answering lots of questions. I am stuck in a similar position. I have read the manuals and I was able to design models and analyse results based on a two factorial level. But having troubles generating results based on multi-factorial (and multi-level) experiments. My experiment is also the same as above and looking for answers to the same questions. It would be great if you could please suggest a similar answer as you have explained everything step wise in a different post for two-factor model design - DESeq2 Model Design . I'm sure lots of us will benefit from this - thanks